Graph Neural Networks in Point Clouds: A Survey

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article reviews the application of graph neural networks to different tasks on point clouds and collates a number of papers. It has some significance. There are the following questions to consider:

1. the format of the charts is not uniform, for example, some tables are three-line tables and some are not, and the charts have a big problem in the layout, please double check and revise.

2. the vein of the text through the task-oriented distinction between different graph convolution method is debatable, why not from the method itself to sort out? If it is task-oriented, is it comprehensive, such as 3D instance segmentation and other tasks?

3. the paper describes the datasets according to different tasks, in fact many datasets are oriented to multiple tasks, is the current classification reasonable?

4. for the datasets in the article, can there be experimental datasets of the last three years, if so, can be added to enhance the persuasiveness of the article.

5. regarding the classification criteria in Figures 3 and 4, is there not a single article for spectral-based methods after 2021, and could the authors please double-check this?

6. The article has elaborated the literature on several tasks, but in general, there are fewer analyses of graphical neural networks on related tasks, such as the advantages and disadvantages of comparing with other methods.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Although there are now many survey papers on the topic, the field is so vast and rapidly evolving that any work attempting to organize and classify the existing research, various trends, and future research directions is welcome. Therefore, the work deserves to be published in its current form.

A few points:

- Given that there are many survey papers, it would be useful to classify them and highlight the contribution of the presented work.

- In the introduction, in the section discussing the number of publications per year consulted, it would be appropriate to provide some information on how the research was conducted (databases consulted, keywords, etc.).

- In all figures showing the timeline of articles, it would be helpful to include the reference number next to the cited method and authors.

- In Figure 4, the timeline does not show any work on spectral-based methods after 2020. If this is not an oversight, as the bibliography seems to suggest, this aspect should be highlighted in the text.

- Line 40: replace the period after "nature" with a comma.

- Figure 1: correct “segmentaion” to “segmentation.”

- Line 258: correct “Base” to “Based.”

- Line 592: there is still a reference to tables 6 and 7, but it should be tables 8 and 9.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors The review is meaningful and comprehensive and may be useful for the readers. However, it need to be improved in the following according to my point of view. 1. More comments might be added to each applications, that is, at the end of each sub-sections in Section 3. This may give readers guidance when they try to choose the better processing method for their specific applications. 2. In the Section 4.2, I suggest add quantitative evaluation of Graph based processing as one of the challenges, because many applications require accurate processing point clouds. 3. In the introduction, point clouds from industrial applications by using the scanner like GOM ATOS have more dense point clouds should be mentioned. 4. Some typos， including but not limited to， （1）in the captions of figure 11 and 12, "form" should be "from". （2）in the title of 3.4.4, "data" should be erased.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

In this study, authors review the applications of GNNs and graph-based methods in point cloud applications based on fundamental tasks, such as segmentation, classification, object detection, registration, and other related tasks. This paper lists the application of mainstream methods in each type of task in chronological order and discusses the development trends and future prospects of graph-based methods. Very interesting results have been provided, and is extremely useful to the science community. However, here are some small mistakes in the paper that need to be checked and corrected. Thus, I would recommend its acceptance for publication after major revision.

1. Introduction: The authors mention the application of GNN in point cloud processing, but fail to fully demonstrate the advantages of GNN over other deep learning models. To strengthen the persuasive power, I suggest that the authors add more comparative analyses, such as quantitative comparisons and qualitative analyses (e.g., interpretability of the model, ability to handle irregular data, etc.), so as to more comprehensively demonstrate the advantages of GNNs in point cloud processing.

2. Section 2. The title of Section 2 is ‘BACKGROUND AND CATEGORIZATION’, but the subtitle is the datasets and evaluation metrics, I think the relationship between them is not significant, authors can reconsider the title in light of the actual content of the section.

3. Section 2.2. The authors list a number of assessment indicators, but do not elaborate on the specific calculations of these assessment indicators. This may lead to confusion for readers in understanding these indicators. 

4. Sections 3.1. It is suggested that a detailed discussion of the advantages and disadvantages of spectral and spatial methods be added by means of comparative analysis and incorporation of practical application cases.

5. The conclusion could be richer by briefly summarizing what has already been stated and by outlining current issues, trends, and perspectives on the future development of the issue.

6. To enhance the contextual coherence of the paper, when describing the current status of each study, the author can briefly elaborate on the interactions and connections between that study and previous or subsequent studies.

7. It is suggested that the authors optimize the graphic layout by placing the figure closest to the citation to enhance the readability and coherence of the article.

8. Figures 1. Correct ‘segmentaion’ to ‘segmentation’.

9. Figures 2. Authors may carefully evaluate the role and necessity of this figure in the paper. If it is felt that the figure is really not necessary, consideration may be given to deleting or replacing it with something else that is more representative.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Good!

Reviewer 3 Report

Comments and Suggestions for Authors

All my conerns in the first round reivew are well addressed. 

Reviewer 4 Report

Comments and Suggestions for Authors

All of my issues have been addressed